Fast CPUs have gotten faster: reduce self-test times by factor 5

[corneliusroemer]

I just downloaded Mlucas on my M1 MBP (32GB RAM) and ran the tiny and small self tests. They were at least a factor of 5 faster than what was stated.

This PR divides all self-test run-times by around a factor of 5.

Self-test runtimes:
tiny: 8s (instead of 1 minute)
small: 1m23s (instead of 10 minutes)

[tdulcet]

Thanks for the PR!

How many threads were you using to benchmark? It looks like you forgot to update the medium and egregious/xxl self-test times.

Note that more of this help documentation file needs to be updated. For example, it is missing the new teensy self-test and a few of other the FFT length ranges were also updated by Ernst in v21, which could partially explain the time discrepancies you noticed. If possible, I would recommend running some of the larger self-tests on your system to confirm the new runtimes.

Mlucas/src/Mlucas.c

Lines 3455 to 3467 in 4db9fbc

	/* Number of distinct FFT lengths supported for self-tests: */
	#define numTest 136 // = sum of all the subranges below
	/* Number of FFT lengths in the various subranges of the full self-test suite: */
	#define numTeensy 15 // v21: added 'Teensy', moved 8 smallest 'Tiny' into it,
	#define numTiny 32 // changed counts from [-,32,32,16,24,16,9,0,0] to [15,32,24,20,20,16,9,0,0]
	#define numSmall 24
	#define numMedium 20
	#define numLarge 20
	#define numHuge 16
	/* Adding larger FFT lengths to test vectors requires supporting changes to Mdata.h:MAX_FFT_LENGTH_IN_K and get_fft_radices.c */
	#define numEgregious 9
	#define numBrobdingnagian 0
	#define numGodzillian 0

[corneliusroemer]

I now also ran the medium (recommended default) test (Mlucas -s m) and it took 10m9s (single CPU), I forgot to update that one actually, done now!

I didn't forget to update the egregious one - I just didn't know whether several days would now be just one day or not, so I played it safe and kept it at several days.